Digital soil mapping in a low-relief landscape to support wetland restoration decisions

Abstract

Efforts to utilize conventional soil maps in wetland conservation and restoration planning are often hampered by the coarse scale of the soil maps relative to the scale of restoration decisions, the spatial aggregation of soil components, and the difficulty in accounting for uncertainty in soil maps. The goal of this study was to explore the potential of digital soil mapping (DSM) techniques to improve identification of wetland soils and map soil properties on a low relief depressional wetland landscape. Separate random forests models were constructed to predict natural soil drainage and texture class on forest and cropland. The models were trained using soil profile data collected from local soil surveyors and previous research. Environmental covariates included topographic metrics developed from a 3 m lidar digital elevation model, and attributes derived from soil survey maps, the agricultural ditch network, and the National Wetlands Inventory. The resulting soil class probability maps demonstrated better representation of soil-landscape relationships on depressional wetlands on forest than on cropland; an independent field validation of soil maps resulted in greater than 70% accuracy in predicting natural soil drainage and texture class on forested depressions. DSM techniques can be used to generate maps with greater spatial detail than conventional soil maps in low relief depressional wetland landscapes; these maps and the attribute importance measures derived from the models have potential to improve watershed models and inform our understanding of wetland hydrology in these landscapes.